In this specification, the term integrated circuit is used to describe a chip or MCM (multi-chip module) embedded with design-for-test (DFT) techniques. The terms circuit assembly and printed circuit board will be considered interchangeable. The term circuit assembly includes printed circuit boards as well as other types of circuit assemblies. A circuit assembly is a combination of integrated circuits. The resulting combination is manufactured to form a physical or functional unit.
An integrated circuit or circuit assembly, in general, contains two or more systems clocks, each controlling one module or logic block, called clock domain. Each system clock is either directly coming from a primary input (edge pin/connector) or generated internally. These system clocks can operate at totally unrelated frequencies (clock speeds), at sub-multiples of each other, at the same frequency but with different clock skews, or at a mix of the above. Due to clock skews among these system clocks, when a DFT technique, such as self-test or scan-test, is employed, it is very likely that faults associated with the function between two clock domains, called crossing clock-domain faults, will become difficult to test. In the worst case, these crossing clock-domain faults when propagating into the receiving clock domain could completely block detection or location of all faults within that clock domain.
Thus, in order to solve the fault propagation problem, DFT approaches are proposed to take over control of all system clocks and reconfigure them as capture clocks.
Prior-art DFT approaches in this area to testing crossing clock-domain faults as well as faults within each clock domain centered on using the isolated DFT, ratio'ed DFT, and one-hot DFT techniques. They are all referred to as single-capture DFT techniques, because none of them can provide multiple skewed capture clocks (or an ordered sequence of capture clocks) in each capture cycle during self-test or scan-test.
In using the isolated DFT technique, all boundary signals crossing a clock domain and flowing into the receiving clock domains are completely blocked or disabled by forcing each of them to a predetermined logic value of 0 or 1. See U.S. Pat. No. 6,327,684 issued to Nadeau-Dostie et al. (2001). This approach, in general, can allow all clock domains to be tested in parallel. The major drawbacks of this approach are that it requires insertion of capture-disabled logic in between clock domains and all scan enable signals each associated with one clock domain must be operated at-speed. The design change could take significant efforts and it might impact normal mode operation. Running all scan enable signals at-speed requires routing them as clock signals using layout clock-tree synthesis (CTS). In addition, since boundary signals can traverse through two clock domains in both directions, this approach requires testing crossing clock-domain faults in two or more test sessions. This could substantially increase the test time required and might make the capture-disabled logic even more complex to implement than anticipated.
In using the ratio'ed DFT technique, all clock domains must be operated at sub-multiples of one reference clock. For instance, assume that a design contains 3 clock domains running at 150 MHz, 80 MHz, and 45 MHz, respectively. The 3 clock domains may have to be operated at 150 MHz, 75 MHz, and 37.5 MHz during testing. See U.S. Pat. No. 5,349,587 issued to Nadeau-Dostie et al. (1994). This approach reduces the complexity of testing a multiple-frequency design and avoids potential races or timing violations crossing clock domains. It can also allow testing of all clock domains in parallel. However, due to changes in clock-domain operating frequencies, this approach loses its self-test or scan-test intent of testing multiple-frequency designs at their rated clock speeds (at-speed) and may require significant design and layout efforts on re-timing (or synchronizing) all clock domains. Power consumption could be also another serious problem because all scan cells (memory elements) are triggered simultaneously every few cycles.
In using the one-hot DFT technique, each crossing clock-domain signal flowing into its receiving clock domains must be initialized to or held at a predetermined logic value of 0 or 1 first. This initialization is usually accomplished by shifting in predetermined logic values to all clock domains so that all crossing clock-domain signals are forced to a known state. Testing is then conducted domain-by-domain, thus, called one-hot testing. See U.S. Pat. No. 5,680,543 issued to Bhawmik et al. (1997). The major benefits of using this approach are that it can still detect or locate crossing clock-domain faults and does not need insertion of disabled logic, in particular, in critical paths crossing clock domains. However, unlike the isolated or ratio'ed DFT approach, this approach requires testing of all clock domains in series, resulting in long test time. It also requires significant design and layout efforts on re-timing (or synchronizing) all clock domains.
Two additional prior-art DFT approaches had also been proposed, one for scan-test, the other for self-test. Both approaches are referred to as multiple-capture DFT techniques, because they can provide multiple skewed capture clocks (or an ordered sequence of capture clocks) in each capture cycle during scan-test or self-test.
The first prior-art multiple-capture DFT approach is to test faults within each clock domain and faults between two clock domains in scan-test mode. See U.S. Pat. No. 6,070,260 issued to Buch et al. (2000) and U.S. Pat. No. 6,195,776 issued to Ruiz et al. (2001). These approaches rest on using multiple skewed scan clocks or multiple skew capture events each operating at the same reduced clock speed in an ATE (automatic test equipment) to detect faults. Combinational ATPG (automatic test pattern generation) is used to generate scan-test patterns and ATE test programs are created to detect faults in the integrated circuit. Unfortunately, currently available ATPG tools only assume the application of one clock pulse (clock cycle) to each clock domain. Thus, these approaches can only detect stuck-at faults in scan-test mode. No prior art using multiple skewed capture clocks were proposed to test delay or stuck-at faults requiring two or more capture clock pulses for full-scan or partial-scan designs.
The second prior-art multiple-capture DFT approach is to test faults within each clock domain and faults between two clock domains in self-test mode. See the paper co-authored by Hetherington et al. (1999). This approach rests on using multiple shift-followed-by-capture clocks each operating at its operating frequency, in a programmable capture window, to detect faults at-speed. It requires clock suppression, complex scan enable (SE) timing waveforms, and shift clock pulses in the capture window to control the capture operation. These shift clock pulses may also need precise timing alignment. As a result, it becomes quite difficult to perform at-speed self-test for designs containing clock domains operated at totally unrelated frequencies, e.g., 133 MHz and 60 MHz.
Thus, there is a need for an improved method, apparatus, or computer-aided design (CAD) system that allows at-speed or slow-speed testing of faults within clock domains and between any two clock domains using a simple multiple-capture DFT technique. The method and apparatus of the present invention will control the multiple-capture operations of the capture clocks in self-test or scan-test mode. It does not require using shift clock pulses in the capture window, inserting capture-disabled logic in normal mode, applying clock suppression on capture clock pulses, and programming complex timing waveforms on scan enable (SE) signals. In addition, the CAD system of the present invention further comprises the computer-implemented steps of performing multiple-capture self-test or scan synthesis, combinational fault simulation, and combinational ATPG that are currently unavailable in the CAD field using multiple-capture DFT techniques.
In light of the prevalent yield loss problem caused by high peak capture power in self-test or scan-test mode, the present invention further comprises methods for capture-power reduction. It also supports scan-chain integrity verification in self-test mode.